Weekly Digest #24: Blurring Faces No Longer Works
September 22nd, 2016
Several groups of scientists simultaneously announced that they are working on blurred image recognition. In the meantime, neural networks are used for conventional predictive input on smartphones and Omega2, the $5 microcomputer, is finally in manufacturing stage, making a smart home a possibility for a wider range of people. This and more - in our digest.
SHORT N’ SWEET
Google launched Allo, the mobile messenger with built-in intelligent voice assistant. This app has been talked about since the Google I/O 2016 conference for developers in spring.
Experts from Massachusetts Institute of Technology have created a new programming language called Milk, designed to work with large data. Tests have shown that the use of this new language will speed up the data processing at least by four times.
Apple launched iMessage App Store. It contains content for the iMessage such as stickers, additional applications, games, and more. At the moment, this is available only to iOS 10 beta users.
According to GitHub, the most active open-source community is Microsoft’s - there are over 16.4 thousand participants in its open-source projects. Second place is taken by Facebook (15.6 thousand participants). Third and fourth place are taken by Docker (14 thousand) and Google (12.1 thousand).
Internet of Things security is one of the problems that more and more experts are talking about. The price for breaking an IoT-network is too high and today it’s even hard to imagine what could happen, which makes it a priority to find a solution. At the recent conference called DEF CON, security problems of various smart devices were discussed, but a few weeks after the conference, new threats were discovered.
Such vulnerabilities were discovered in smart door locks, manufactured by QuickLock, iBlulock, Plantraco, Ceomate, Elecycle, Vians, Lagute, Okidokeys, Danalock, and other. These were found to be vulnerable to password interception and other cyber attacks.
Overall, over a hundred critical errors in IoT-devices were identified at the DEF CON. The participants stressed that many manufacturers pay insufficient attention to security issues, and even ignore experts’ recommendations sometimes.
Omega2: Internet of Things for $5
There is a belief that smart home is a very complex and expensive process. This view is being challenged by developers who create simple and affordable devices, a special place among which is occupied by single-board computers.
After successfully finishing the Kickstarter fundraising (we wrote about it earlier here), the manufacturing process has begun. The gadget is designed for building an IoT-network at home. In size, it’s no bigger than an ordinary postage stamp, but it is equipped with Wi-Fi and Flash memory. Using this tiny computer, one can program in the usual daily tasks such as turning on and off the lights, audio center, home control system, climate control, and more. Authors of Omega2 project position it as the easiest way to automate your home and make Internet of Things into an everyday technology.
Omega2 is several times smaller than its popular competitors - Arduino and Raspberry Pi. Furthermore, it is at least five times cheaper than these models of SBCs. Omega2 has a built-in development tool for C++, Python, PHP, Node.js, and Ruby programming and is integrated with Onion Cloud file storage.
WELT - Belt that counts your steps and protects from overeating
Fitness these days can take quite fancy embodiments. No one is excited about fitness bracelets anymore - they are everywhere. The authors of WELT smart belt decided to take another path and combine fitness tracking with a familiar item of clothing.
The belt can count steps and notify users if their waist size is increasing. This is done by analyzing the belt tension. If the person ate too much, it would increase the waist, the belt would stretch notify the user that he shouldn’t have consumed that many calories.
Another interesting feature of the smart belt is the ability to remind the person if he was motionless for some time, for example, if the person was sitting in front of the computer or elsewhere.
The analysis of data received from the belt is done via a special app (available for Android and iOS). The belt is equipped with a battery that supports micro-USB.
The new version of Swift 3.0
Apple has introduced a new version of Swift 3.0 programming language. This is the first upgrade since the original code became open-source in December 2015. At the moment, the official assembly is prepared for Linux (Ubuntu 14.04, 15.10) and macOS (Xcode).
The new release includes previously unavailable for Linux libraries set called Swift Core Libraries. The Linux version is not linked to the Objective-C Runtime, which means you can use the new language in an environment that doesn’t support Objective-C, however, it is incompatible with previous versions of the language.
Blurring Images made useless by Neural network
Blurring images is the easiest way to protect the photos you have to make available for the public. For example, Google blurs the photos of its Google Street View and Google Maps services where there are images of people’s faces. Media people often blur images of children. However, in a short time, such tactic will not make sense anymore because two groups of scientists have reported that they have learned to overcome such blur.
Scientists from the Max Planck Institute (Germany) have taught a neural network to recognize distorted images. Faceless Recognition System has learned to identify blurry faces in 91.5% of times. Even if you cover a person’s face with a black square, Faceless Recognition System will be able to know who the person is in 47% of times. For the first time, machine learning was implemented as a solution for the image recognition. Prior to this, Facebook has shared about the algorithm used for facial recognition in the social network that is able to identify faces with 83% accuracy.
Another group of scientists from the University of Texas has created a machine learning system that is also able to recognize blurry faces. These experts have used an open-source software platform called Torch and face recognition algorithms to train a neural network. The resulting system can recognize a person with an accuracy of 90%. Such accuracy is obtained in the image analysis that used YouTube pixelization tools. Graphics editors’ blurriness can be recognized in 50-70% of cases and encryption tool P3 (Privacy-Preserving Photo Sharing) gives the best protection to one’s photos and the images could be recognized only in 17% of cases.
SwiftKey, mobile keyboard, gets a neural network support
Neural networks can be used not only to solve complex problems, such as financial forecasting and pattern recognition. Developers decided to implement neural network algorithms into a new version of the mobile keyboard called SwiftKey and use it for predictive input.
Initially, the developers have released SwiftKey Neural Alpha standalone application as an experiment, but eventually moved the improved and tested algorithms to the main app.
The neural network used in SwiftKey tries to understand what the user is saying and offers him the most suitable options to continue. The developers emphasize that the new technology, unlike the previous one, analyzes the meaning of the phrase before offering word options. Another great feature is that the new technology works locally on the device and doesn’t require a connection to the cloud. At the moment, SwiftKey can only work with English version and the neural version is only available for Android.